Imputation Using SUDAAN PROC HOTDECK for Education Surveys
Creel, D. V. (2010, May). Imputation Using SUDAAN PROC HOTDECK for Education Surveys. Presented at AAPOR 2010, .
Most surveys implement methods to reduce nonresponse, unfortunately almost all surveys experience some level of nonresponse at the unit and item level. In this paper, we focus on the potential impact of item nonresponse on survey estimates and, primarily, on using the new HOTDECK procedure from SUDAAN® 10.0 to address item nonresponse. One method of analyzing data with item nonresponse is complete case analysis. That is, only analyze the data that do not have any missing values related to the analysis. One problem with complete case analysis is that it ignores any information that the item nonrespondents may have which creates the potential for biased survey estimates, if the missingness is not missing completely at random.
An alternative to complete case analysis is to fill in or impute the missing items. Analyzing data with imputed values has the advantage of keeping all of the data available for analysis and, hopefully, minimizing the potential for biased survey estimates. Although imputation can be a resource intensive task, one way to reduce the resources allocated to the imputation task is to use flexible standardized software. SUDAAN® 10.0 has introduced a procedure called PROC HOTDECK which implements the weighted sequential hot deck imputation methodology. We demonstrate the application of PROC HOTDECK to impute a single variable on an education survey, to simultaneously impute multiple variables on an education survey, and to multiply impute through Monte Carlo simulation.